Version: 2.2.1

Storm MQTT Integration

About

MQTT is a lightweight publish/subscribe protocol frequently used in IoT applications.

Further information can be found at http://mqtt.org. The HiveMQ website has a great series on MQTT Essentials.

Features include:

  • Full MQTT support (e.g. last will, QoS 0-2, retain, etc.)
  • Spout implementation(s) for subscribing to MQTT topics
  • A bolt implementation for publishing MQTT messages
  • A trident function implementation for publishing MQTT messages
  • Authentication and TLS/SSL support
  • User-defined "mappers" for converting MQTT messages to tuples (subscribers)
  • User-defined "mappers" for converting tuples to MQTT messages (publishers)

Quick Start

To quickly see MQTT integration in action, follow the instructions below.

Start a MQTT broker and publisher

The command below will create an MQTT broker on port 1883, and start a publsher that will publish random temperature/humidity values to an MQTT topic.

Open a terminal and execute the following command (change the path as necessary):

java -cp examples/target/storm-mqtt-examples-*-SNAPSHOT.jar org.apache.storm.mqtt.examples.MqttBrokerPublisher

Run the example toplogy

Run the sample topology using Flux. This will start a local mode cluster and topology that consists of the MQTT Spout publishing to a bolt that simply logs the information it receives.

In a separate terminal, run the following command (Note that the storm executable must be on your PATH):

storm jar ./examples/target/storm-mqtt-examples-*-SNAPSHOT.jar org.apache.storm.flux.Flux ./examples/src/main/flux/sample.yaml --local

You should see data from MQTT being logged by the bolt:

27020 [Thread-17-log-executor[3 3]] INFO  o.a.s.f.w.b.LogInfoBolt - {user=tgoetz, deviceId=1234, location=office, temperature=67.0, humidity=65.0}
27030 [Thread-17-log-executor[3 3]] INFO  o.a.s.f.w.b.LogInfoBolt - {user=tgoetz, deviceId=1234, location=office, temperature=47.0, humidity=85.0}
27040 [Thread-17-log-executor[3 3]] INFO  o.a.s.f.w.b.LogInfoBolt - {user=tgoetz, deviceId=1234, location=office, temperature=69.0, humidity=94.0}
27049 [Thread-17-log-executor[3 3]] INFO  o.a.s.f.w.b.LogInfoBolt - {user=tgoetz, deviceId=1234, location=office, temperature=4.0, humidity=98.0}
27059 [Thread-17-log-executor[3 3]] INFO  o.a.s.f.w.b.LogInfoBolt - {user=tgoetz, deviceId=1234, location=office, temperature=51.0, humidity=12.0}
27069 [Thread-17-log-executor[3 3]] INFO  o.a.s.f.w.b.LogInfoBolt - {user=tgoetz, deviceId=1234, location=office, temperature=27.0, humidity=65.0}

Either allow the local cluster to exit, or stop it by typing Cntrl-C.

MQTT Fault Tolerance In Action

After the toplogy has been shutdown, the MQTT subscription created by the MQTT spout will persist with the broker, and it will continue to receive and queue messages (as long as the broker is running).

If you run the toplogy again (while the broker is still running), when the spout initially connects to the MQTT broker, it will receive all the messages it missed while it was down. You should see this as burst of messages, followed by a rate of about two messages per second.

This happens because, by default, the MQTT Spout creates a session when it subscribes -- that means it requests that the broker hold onto and redeliver any messages it missed while offline. Another important factor is the the MqttBrokerPublisher publishes messages with a MQTT QoS of 1, meaning at least once delivery.

For more information about MQTT fault tolerance, see the Delivery Guarantees section below.

Delivery Guarantees

In Storm terms, the MQTT Spout provides at least once delivery, depending on the configuration of the publisher as well as the MQTT spout.

The MQTT protocol defines the following QoS levels:

  • 0 - At Most Once (AKA "Fire and Forget")
  • 1 - At Least Once
  • 2 - Exactly Once

This can be a little confusing as the MQTT protocol specification does not really address the concept of a node being completely incinerated by a catasrophic event. This is in stark contrast with Storm's reliability model, which expects and embraces the concept of node failure.

So resiliancy is ultimately dependent on the underlying MQTT implementation and infrastructure.

Recommendations

You will never get at exactly once processing with this spout. It can be used with Trident, but it won't provide transational semantics. You will only get at least once guarantees.

If you need reliability guarantees (i.e. at least once processing):

  1. For MQTT publishers (outside of Storm), publish messages with a QoS of 1 so the broker saves messages if/when the spout is offline.
  2. Use the spout defaults (cleanSession = false and qos = 1)
  3. If you can, make sure any result of receiving and MQTT message is idempotent.
  4. Make sure your MQTT brokers don't die or get isolated due to a network partition. Be prepared for natural and man-made disasters and network partitions. Incineration and destruction happens.

Configuration

For the full range of configuration options, see the JavaDoc for org.apache.storm.mqtt.common.MqttOptions.

Message Mappers

To define how MQTT messages are mapped to Storm tuples, you configure the MQTT spout with an implementation of the org.apache.storm.mqtt.MqttMessageMapper interface, which looks like this:

public interface MqttMessageMapper extends Serializable {

    Values toValues(MqttMessage message);

    Fields outputFields();
}

The MqttMessage class contains the topic to which the message was published (String) and the message payload (byte[]). For example, here is a MqttMessageMapper implementation that produces tuples based on the content of both the message topic and payload:

/**
 * Given a topic name: "users/{user}/{location}/{deviceId}"
 * and a payload of "{temperature}/{humidity}"
 * emits a tuple containing user(String), deviceId(String), location(String), temperature(float), humidity(float)
 *
 */
public class CustomMessageMapper implements MqttMessageMapper {
    private static final Logger LOG = LoggerFactory.getLogger(CustomMessageMapper.class);


    public Values toValues(MqttMessage message) {
        String topic = message.getTopic();
        String[] topicElements = topic.split("/");
        String[] payloadElements = new String(message.getMessage()).split("/");

        return new Values(topicElements[2], topicElements[4], topicElements[3], Float.parseFloat(payloadElements[0]), 
                Float.parseFloat(payloadElements[1]));
    }

    public Fields outputFields() {
        return new Fields("user", "deviceId", "location", "temperature", "humidity");
    }
}

Tuple Mappers

When publishing MQTT messages with the MQTT bolt or Trident function, you need to map Storm tuple data to MQTT messages (topic/payload). This is done by implementing the org.apache.storm.mqtt.MqttTupleMapper interface:

public interface MqttTupleMapper extends Serializable{

    MqttMessage toMessage(ITuple tuple);

}

For example, a simple MqttTupleMapper implementation might look like this:

public class MyTupleMapper implements MqttTupleMapper {
    public MqttMessage toMessage(ITuple tuple) {
        String topic = "users/" + tuple.getStringByField("userId") + "/" + tuple.getStringByField("device");
        byte[] payload = tuple.getStringByField("message").getBytes();
        return new MqttMessage(topic, payload);
    }
}

MQTT Spout Parallelism

It's recommended that you use a parallelism of 1 for the MQTT spout, otherwise you will end up with multiple instances of the spout subscribed to the same topic(s), resulting in duplicate messages.

If you want to parallelize the spout, it's recommended that you use multiple instances of the spout in your topolgoy and use MQTT topic selectors to parition the data. How you implement the partitioning strategy is ultimately determined by your MQTT topic structure. As an example, if you had topics partitioned by region (e.g. east/west) you could do something like the following:

String spout1Topic = "users/east/#";
String spout2Topic = "users/west/#";

and then join the resulting streams together by subscribing a bolt to each stream.

Using Flux

The following Flux YAML configuration creates the toplolgy used in the example:

name: "mqtt-topology"

components:
   ########## MQTT Spout Config ############
  - id: "mqtt-type"
    className: "org.apache.storm.mqtt.examples.CustomMessageMapper"

  - id: "mqtt-options"
    className: "org.apache.storm.mqtt.common.MqttOptions"
    properties:
      - name: "url"
        value: "tcp://localhost:1883"
      - name: "topics"
        value:
          - "/users/tgoetz/#"

# topology configuration
config:
  topology.workers: 1
  topology.max.spout.pending: 1000

# spout definitions
spouts:
  - id: "mqtt-spout"
    className: "org.apache.storm.mqtt.spout.MqttSpout"
    constructorArgs:
      - ref: "mqtt-type"
      - ref: "mqtt-options"
    parallelism: 1

# bolt definitions
bolts:
  - id: "log"
    className: "org.apache.storm.flux.wrappers.bolts.LogInfoBolt"
    parallelism: 1


streams:
  - from: "mqtt-spout"
    to: "log"
    grouping:
      type: SHUFFLE

Using Java

Similarly, you can create the same topology using the Storm Core Java API:

TopologyBuilder builder = new TopologyBuilder();
MqttOptions options = new MqttOptions();
options.setTopics(Arrays.asList("/users/tgoetz/#"));
options.setCleanConnection(false);
MqttSpout spout = new MqttSpout(new StringMessageMapper(), options);

MqttBolt bolt = new LogInfoBolt();

builder.setSpout("mqtt-spout", spout);
builder.setBolt("log-bolt", bolt).shuffleGrouping("mqtt-spout");

return builder.createTopology();

SSL/TLS

If the MQTT broker you are connecting to requires SSL or SSL client authentication, you need to configure the spout with an appropriate URI, and the location of keystore/truststore files containing the necessary certificates.

SSL/TLS URIs

To connect over SSL/TLS use a URI with a prefix of ssl:// or tls:// instead of tcp://. For further control over the algorithm, you can specify a specific protocol:

  • ssl:// Use the JVM default version of the SSL protocol.
  • sslv*:// Use a specific version of the SSL protocol, where * is replaced by the version (e.g. sslv3://)
  • tls:// Use the JVM default version of the TLS protocol.
  • tlsv*:// Use a specific version of the TLS protocol, where * is replaced by the version (e.g. tlsv1.1://)

Specifying Keystore/Truststore Locations

The MqttSpout, MqttBolt and MqttPublishFunction all have constructors that take a KeyStoreLoader instance that is used to load the certificates required for TLS/SSL connections. For example:

 public MqttSpout(MqttMessageMapper type, MqttOptions options, KeyStoreLoader keyStoreLoader)

The DefaultKeyStoreLoader class can be used to load certificates from the local filesystem. Note that the keystore/truststore need to be available on all worker nodes where the spout/bolt might be executed. To use DefaultKeyStoreLoader you specify the location of the keystore/truststore file(s), and set the necessary passwords:

DefaultKeyStoreLoader ksl = new DefaultKeyStoreLoader("/path/to/keystore.jks", "/path/to/truststore.jks");
ksl.setKeyStorePassword("password");
ksl.setTrustStorePassword("password");
//...

If your keystore/truststore certificates are stored in a single file, you can use the one-argument constructor:

DefaultKeyStoreLoader ksl = new DefaultKeyStoreLoader("/path/to/keystore.jks");
ksl.setKeyStorePassword("password");
//...

SSL/TLS can also be configured using Flux:

name: "mqtt-topology"

components:
   ########## MQTT Spout Config ############
  - id: "mqtt-type"
    className: "org.apache.storm.mqtt.examples.CustomMessageMapper"

  - id: "keystore-loader"
    className: "org.apache.storm.mqtt.ssl.DefaultKeyStoreLoader"
    constructorArgs:
      - "keystore.jks"
      - "truststore.jks"
    properties:
      - name: "keyPassword"
        value: "password"
      - name: "keyStorePassword"
        value: "password"
      - name: "trustStorePassword"
        value: "password"

  - id: "mqtt-options"
    className: "org.apache.storm.mqtt.common.MqttOptions"
    properties:
      - name: "url"
        value: "ssl://raspberrypi.local:8883"
      - name: "topics"
        value:
          - "/users/tgoetz/#"

# topology configuration
config:
  topology.workers: 1
  topology.max.spout.pending: 1000

# spout definitions
spouts:
  - id: "mqtt-spout"
    className: "org.apache.storm.mqtt.spout.MqttSpout"
    constructorArgs:
      - ref: "mqtt-type"
      - ref: "mqtt-options"
      - ref: "keystore-loader"
    parallelism: 1

# bolt definitions
bolts:

  - id: "log"
    className: "org.apache.storm.flux.wrappers.bolts.LogInfoBolt"
    parallelism: 1


streams:

  - from: "mqtt-spout"
    to: "log"
    grouping:
      type: SHUFFLE

Committer Sponsors